Firefox's Smart Window AI now runs on Mistral Small 4, an open model
The browser's default assistant is Apache-2.0 and self-hostable, keeps zero data retention, and just expanded to France — plus Gemini 3.8 Live opens on Vercel.
Copy markdown
Firefox's new assistant runs Mistral Small 4
Firefox Smart Window beta now uses Mistral Small 4 to summarize messy search trails, resurface a page you clicked away from, and pull answers straight from your open tabs. It's live in the US, Canada, and now France with native French support; the UK, Germany, and more of Europe follow later in 2026.
The model is Apache-2.0 — run it yourself
Mistral Small 4 is a 119B open-weight model that folds reasoning, vision, and coding into one checkpoint, already on Hugging Face under Apache 2.0 with GGUF quants posted for local runs. The exact assistant powering a mainstream browser is one you can download and self-host behind your own private tooling.
Zero data retention, and the model stays swappable
Smart Window conversations aren't saved on Mozilla's servers by default, and Firefox still lets users choose other AI providers instead of locking to one. If you build browser-adjacent or privacy-sensitive features, it's a shipping example of a provider-neutral, no-retention default.
Elsewhere: Gemini 3.8 Live opens on Vercel AI Gateway
Google's real-time audio models — gemini-3.8-live and its extended-thinking variant — are now callable through Vercel AI Gateway, so you can wire a low-latency voice agent without standing up direct Google API access.